This is ChatGPT, created by Stephen Wolfram, the world's top AI scholar, a major change-maker in the field of science and technology, and the creator of WolframAlpha, the world's first practical artificial intelligence search engine, as an artificial intelligence chatbot program developed by OpenAI, ChatGPT was unveiled in November 2022, and its automatically generated text is amazing, as if written by humans. This book will take you inside the world of ChatGPT and reveal how it produces human-like texts.
>> Featured Content:
The essence of ChatGPT lies in its simplicity and directness: it extracts a large number of human text samples from the Internet, books, and other sources, and trains neural networks to generate "similar" text. In particular, ChatGPT can start with a given "prompt" and continue to generate text similar to the training data.
The role of neural networks in ChatGPT is fairly basic, although it is made up of billions of simple elements. Whenever the neural network generates a new word (or part of a word), it passes it to all elements in turn based on the input obtained from the currently generated text (no loops, etc.).
Surprisingly, this process successfully produces text that is similar to content found on the Internet, books, and other sources. ChatGPT not only generates coherent human language, but also "follows the prompts" based on what has been "read". But it doesn't always produce meaningful conversations in a global sense, because (in the absence of Wolfram|With the help of Alpha's computing power) it is simply imitating what is in the training material, trying to "say the right thing" as much as possible.
The implementation of ChatGPT is undoubtedly remarkable. However, ChatGPT simply extracts some "coherent textual threads" from its accumulated "statistics of conventional wisdom." Still, the results have reached a surprising level of humanity.
This reveals an important fact in science: human language and the thinking patterns behind it are structurally simpler and more regular than we think. ChatGPT already hints at this, but we can clearly reveal it through tools such as semantic grammar, computational linguistics, etc.
ChatGPT does an extremely good job of generating text, and the results are often very similar to human-created text. Does this imply that ChatGPT works similarly to the human brain? The underlying artificial neural network structure is actually a model of the idealized brain. In the process of human generation of language, many aspects seem to be extremely similar.
However, when it comes to training (or learning), differences in hardware (and some untapped underlying algorithmic ideas) between the brain and modern computers force ChatGPT to adopt a strategy that may be completely different (and in some ways less efficient) than the brain.
It is worth mentioning that there are no "loops" or "recalculation of data" inside ChatGPT. This undoubtedly limits its computing power – even compared to modern computers, let alone the brain.
We don't know how to solve this problem while maintaining the efficiency of the system's training. But if it does, future ChatGPT could do more "brain-like things." Of course, there are many things the brain is not good at, especially when it comes to irreducible calculations. For these problems, both the brain and systems like ChatGPT need to seek the help of "external tools," such as the Wolfram Language.
However, it's very exciting to see what ChatGPT has been able to do at the moment. In a way, it's an excellent example of how a lot of simple computational elements can do amazing things. It also provides us with an optimal motivation to better understand the core human identity, human language, and the nature and principles of the thought processes behind it.